Attention-Based Learning on Molecular Ensembles

TL;DR

This paper introduces an end-to-end deep learning method that directly processes small-molecule conformational ensembles using graph neural networks and attention mechanisms to improve virtual screening based on 3D molecular geometry.

Contribution

The work presents a novel set-based learning approach combining graph neural networks and attention mechanisms for analyzing molecular conformations in virtual screening.

Findings

Feasibility demonstrated on biaryl ligand coordination task.

Attention pooling elucidates key conformational poses.

Potential for improved small molecule virtual screening.

Abstract

The three-dimensional shape and conformation of small-molecule ligands are critical for biomolecular recognition, yet encoding 3D geometry has not improved ligand-based virtual screening approaches. We describe an end-to-end deep learning approach that operates directly on small-molecule conformational ensembles and identifies key conformational poses of small-molecules. Our networks leverage two levels of representation learning: 1) individual conformers are first encoded as spatial graphs using a graph neural network, and 2) sampled conformational ensembles are represented as sets using an attention mechanism to aggregate over individual instances. We demonstrate the feasibility of this approach on a simple task based on bidentate coordination of biaryl ligands, and show how attention-based pooling can elucidate key conformational poses in tasks based on molecular geometry. This work illustrates how set-based learning approaches may be further developed for small molecule-based virtual screening.